Featured White Papers
- Don't miss this enterprise mobility Webcast! (TechRepublic)
- Hosted CRM comparison guide (Inside CRM)
- Enterprise PBX buyer's guide (VoIP-News)
Content provided in partnership with
Health Care Industry
Industry: Email Alert RSS FeedApplication of a propensity score approach for risk adjustment in profiling multiple physician groups on asthma care
Health Services Research, Feb, 2005 by I-Chan Huang, Constantine Frangakis, Francesca Dominici, Gregory B. Diette, Albert W. Wu
Table A1: Propensity Scores of Patients i Enrolling in Each
of 20 Physician Groups
The ith Patient
The jth group 1 2 3
1 [e.sub.1], 1 [e.sub.2], 1 [e.sub.3], 1
2 [e.sub.1], 2 [e.sub.2], 2 [e.sub.3], 2
*
j
*
20 [e.sub.1], 20 [e.sub.2], 20 [e.sub.3], 20
The jth group 4 5 6
1 [e.sub.4], 1 [e.sub.5], 1 [e.sub.6], 1
2 [e.sub.4], 2 [e.sub.5], 2 [e.sub.6], 2
*
j
*
20 [e.sub.4], 20 [e.sub.5], 20 [e.sub.6], 20
The jth group 7 ... i ...
1 [e.sub.7], 1 ... * ...
2 [e.sub.7], 2 *
* *
j [e.sub.i], j
* *
20 [e.sub.7], 20 ... * ...
The jth group 2,515
1 [e.sub.2,515], 1
2 [e.sub.2,515], 2
*
j
*
20 [e.sub.2,515], 3
Table A2: Comparing the jth Physician Group versus
Other 19 Physician Groups
Patients Actually in the
jth Physician Group
(j = 1, ..., 20)
Patients Who Are
Propensity Satisfied with Total Patients
Stratum (s) Care in Strata (s) in Strata (s)
(1) (2) (3)
1 [C.sub.j,1] [N.sub.j,1]
2 [C.sub.j,2] [N.sub.j,2]
3 [C.sub.j,3] [N.sub.j,3]
4 [C.sub.j,4] [N.sub.j,4]
5 [C.sub.j,5] [N.sub.j,5]
Overall [SIGMA] [C.sub.j,s] [SIGMA] [N.sub.j,s]
Patients Actually
in the Other 19
Physician Groups Patients in
(Exclude the 20 Physician Groups
jth Group)
Propensity Total Patients Total Patients
Stratum (s) in Strata (s) in Strata (s)
(1) (4) (5)
1 [M.sub.j,1] [N.sub.j,1] + [M.sub.j,1]
2 [M.sub.j,2] [N.sub.j,2] + [M.sub.j,2]
3 [M.sub.j,3] [N.sub.j,3] + [M.sub.j,3]
4 [M.sub.j,4] [N.sub.j,4] + [M.sub.j,4]
5 [M.sub.j,5] [N.sub.j,5] + [M.sub.j,5]
Overall [SIGMA] [M.sub.j,s] [SIGMA] [N.sub.j,s] + [M.sub.j,s]
Table 1: Analytic Framework for Comparing Risk-Adjustment
Methods for Physician Group Profiling
Risk-Adjustment Method
Method Description Risk Adjustor
Method 1 Hierarchical outcome Sociodemographic
regression adjustment (age, sex, education
without propensity level, types of
scores insurance, drug
coverage), Clinical
(asthma severity and
number of co-morbid
conditions), Health
status (SF-36
physical and mental
component scores)
Method 2 Propensity score- Same as for Method 1
based risk adjustment
Risk-Adjustment Method
Method Remarks
Method 1 1. Adjusts for covariate
effects on patient
satisfaction
2. Addresses
regression-to-the-
mean using
hierarchical
regression on the
covariates
Method 2 1. Adjusts for covariate
effects on provider
selection, using
propensity scores;
does not adjust for
effects on satisfaction
2. Addresses regression-
to-the-mean using
shrinkage techniques *
on the propensity-score
based proportions of
satisfaction
* Using Morris's approach (Morris 1983).
Table 2: Characteristics of Patients with Asthma (n = 2,515)
Dimension Frequency or Mean (SD)
Age (%)
18-24 7.2
25-34 22.0
35-44 34.6
45-54 33.2
55 and above 3.1
Overall, mean (SD) 39.9 (9.5)
Sex (%)
Male 28.8
Female 71.2
Race (%)
White 70.3
African American 5.1
Asian American 10.0
Other 14.7
Education (%)
High school or below 18.4
College 65.3
Graduate 16.3
Health insurance status (%)
Private--through employer 69.1
Private--through self-purchase 24.8
Public--Medicare, Medicaid 1.4
Other 4.9
Drug insurance coverage (%) 96.5
Asthma severity (%)
Mild intermittent 14.4
Mild persistent 19.2
Moderate persistent 49.3
Severe persistent 17.1
Number of comorbidity, mean (SD) 2.1 (1.4)
SF-36 Physical component score, mean (SD) 45.7 (10.3)
SF-36 Mental component score, mean (SD) 47.4 (10.7)
Satisfaction with asthma care
More satisfied with asthma care 55.4
Less satisfied with asthma care 44.7
Table 3: Performance of 20 Physician Groups Estimated Using
Different Methods
No Risk Adjustment
Number of
Patients Unadjusted
Group ID in Group Rate (%) OR (SE)
1 * 163 63.8 1.0
2 177 60.5 0.87 (0.19)
3 151 58.3 0.79 (0.18)
4 212 59.0 0.82 (0.17)
5 63 71.4 1.42 (0.46)
6 86 59.3 0.83 (0.23)
7 146 49.3 0.55 (0.13)
8 82 58.5 0.80 (0.22)
9 110 78.2 2.03 (0.57)
10 75 53.3 0.65 (0.18)
11 64 37.5 0.34 (0.10)
12 103 47.6 0.51 (0.13)
13 176 48.9 0.54 (0.12)
14 141 36.9 0.33 (0.08)
15 31 38.7 0.36 (0.14)
16 164 61.6 0.91 (0.21)
17 194 48.5 0.53 (0.12)
18 110 50.9 0.59 (0.15)
19 218 58.7 0.81 (0.17)
20 49 49.0 0.54 (0.18)
Hierarchical Outcome Propensity Score-Based
Regression Adjustment Risk Adjustment
Adjusted Adjusted
Group ID Rates (%) OR (SE) Rates (%) OR (SE)
1 * 64.7 1.0 57.7 1.0
2 65.4 1.03 (0.23) 63.7 1.29 (0.68)
3 62.2 0.90 (0.21) 54.5 0.88 (0.24)
4 65.8 1.05 (0.23) 57.9 1.01 (0.34)
5 68.9 1.21 (0.32) 61.6 1.18 (0.52)
6 59.4 0.80 (0.21) 51.2 0.77 (0.20)
7 58.5 0.77 (0.18) 50.5 0.75 (0.19)
8 64.2 0.98 (0.25) 60.8 1.14 (0.48)
9 76.4 1.77 (0.45) 67.7 1.54 (1.05)
10 62.0 0.89 (0.23) 54.2 0.87 (0.23)
11 52.8 0.61 (0.17) 42.3 0.54 (0.20)
12 55.1 0.67 (0.17) 46.1 0.63 (0.19)
13 57.9 0.75 (0.17) 50.8 0.76 (0.20)
14 49.2 0.53 (0.13) 39.0 0.47 (0.22)
15 56.0 0.71 (0.21) 52.4 0.81 (0.21)
16 65.1 1.02 (0.24) 59.0 1.06 (0.40)
17 57.5 0.74 (0.17) 53.6 0.85 (0.22)
18 60.9 0.85 (0.21) 52.7 0.82 (0.21)
19 66.0 1.06 (0.23) 59.3 1.07 (0.40)
20 59.1 0.79 (0.22) 55.0 0.90 (0.24)
* Physician group 1 as the reference group.
OR = odds ratio.
Table 4: Characteristics of Physician Groups That Shifted Quintile
Rankings Based on Different Risk Adjustment Methods *
Number
ID # of of
Physician Patients Patient Characteristics
Group Location in Group ([dagger])
(A) Raising ranks after using propensity score method
2 Northern California 177 Gender, severity, number
of comorbidity, drug
prescription coverage,
PCS, MCS
8 Northern California 82 Age, gender, severity,
number of comorbidity,
drug prescription
coverage, PCS, MCS
15 Southern California 31 Age, gender, severity,
number of comorbidity,
drug prescription
coverage, PCS, MCS
17 Southern California 194 Age, gender, PCs, MCS
20 Southern California 49 Age, gender, severity,
number of comorbidity,
PCS, MCS
(B) Lowering ranks after using propensity score method
4 Northen California 212 Age, gender, severity,
number of comorbidity,
PCS
6 Northen California 86 Age, gender, severity,
number of comorbidity,
drug prescription
coverage, PCS, MCS
7 Northern California 146 Age, gender, number of
comorbidity, drug
prescription coverage
18 Southern California 110 Age, gender, severity,
number of comorbidity,
drug prescription
coverage, PCS
19 Southern California 218 Age, gender, drug
prescription coverage,
PCS, MCS
* Method 1 (hierarchical outcome regression adjustment without using
propensity score), Method 2 (propensity score-based risk adjustment).
([dagger]) Statistically different from grand mean (p< 0.05).
PCS = physical component score; MCS = mental component score.
Find Research Guides for:
Find Featured Titles for: Home & Garden
Most Popular Articles in Health
- Fuel your workout: exercisers who eat before they work out have more energy ...
- Soothe a dry, itchy scalp: 5 easy expert solutions
- Cocktails and calories: Beer, wine and liquor calories can really add up. ...
- The sour truth about apple cider vinegar - evaluation of therapeutic use
- The, six best supplements you've never heard of: these secret weapons can ...
- More »